Question Answering on TruthfulQA (MC1, Δ%)
27.54MC1 AccuracyDOLA dynamic
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DOLA dynamicModel=Qwen2.5-7B2026.06 | 27.54 | 26.4 | |
| CRYSTALBOOST (best)Model=Qwen2.5-7B2026.06 | 22.28 | 2.2 | |
| BaselineModel=Qwen2.5-7B2026.06 | 21.79 | — | |
| CRYSTALBOOST (best)Model=Llama-3.1-8B2026.06 | 20.56 | 8.4 | |
| DOLA dynamicModel=Llama-3.1-8B2026.06 | 19.34 | 1.9 | |
| BaselineModel=Llama-3.1-8B2026.06 | 18.97 | — |